Running Head: BENCHMARK EFFECTS
Benchmark Eye Movement Effects during Natural Reading in Autism Spectrum
Disorder.
Philippa L. Howard*, Simon P. Liversedge & Valerie Benson
University of Southampton
*Corresponding author: [email protected]
Psychology
Shackleton Building (B44)
Highfield Campus
University of Southampton
Southampton
SO17 1BJ
1
BENCHMARK EFFECTS
2
Abstract
In two experiments, eye tracking methodology was used to assess online lexical, syntactic
and semantic processing in autism spectrum disorder (ASD). In Experiment 1, lexical
identification was examined by manipulating the frequency of target words. Both typically
developed (TD) and ASD readers showed normal frequency effects, suggesting that the
processes TD and ASD readers engage in to identify words are comparable. In Experiment 2,
syntactic parsing and semantic interpretation requiring the online use of world knowledge
were examined, by having participants read garden path sentences containing an ambiguous
prepositional phrase. Both groups showed normal garden path effects when reading low
attached sentences and the time course of reading disruption was comparable between
groups. This suggests that not only do ASD readers hold similar syntactic preferences to TD
readers, but also that they use world knowledge online during reading. Together, these
experiments demonstrate that the initial construction of sentence interpretation appears to be
intact in ASD. However, the finding that ASD readers skip target words less often in
Experiment 2, and take longer to read sentences during second pass for both experiments,
suggests that they adopt a more cautious reading strategy and take longer to evaluate their
sentence interpretation prior to making a manual response.
Key phrases: Autism Spectrum Disorder, Reading, Eye Movements, Lexical Identification,
Syntactic Ambiguity Resolution.
BENCHMARK EFFECTS
3
Benchmark Eye Movement Effects during Natural Reading in Autism Spectrum Disorder.
Autism spectrum disorder (ASD) is characterised by restrictive and repetitive patterns
of behaviour and significant impairments in social interaction/communication (American
Psychiatric Association, 2013). The cognitive differences that underpin this behavioural
phenotype are often found to affect reading ability (e.g. Nation, Clarke, Wright, & Williams,
2006). Individuals with ASD who do not have an associated learning difficulty or those who
have Asperger syndrome, in general, are found to have intact performance accuracy for ‘low
level’, basic reading tasks (e.g. word identification; Frith & Snowling, 1983; Huemer &
Mann, 2010; Mayes & Calhoun, 2006; Minshew, Goldstein, & Siegel, 1995; Saldaña,
Carreiras, & Frith, 2009), but frequently display impairments in performance for ‘higher
order’ reading tasks (e.g. text comprehension and inferencing; Huemer & Mann, 2010;
Jolliffe & Baron-Cohen, 1999, 2000; Jones et al. 2009; Minshew et al. 1995; but see also
Asberg, Kopp, Berg-Kelly, & Gillberg, 2010).
As yet, there has been no specific theoretical explanation put forward to explain how
linguistic processing occurs in ASD. Previous hypotheses as to why performance on higher
order linguistic tasks is impaired have been derived from domain general cognitive accounts
of ASD. The Weak Central Coherence Theory (WCC) claims ASD to result in a local
processing bias coupled with a lack of spontaneous strive for global coherence (Frith, 1989;
Frith & Happé, 1994; Happé & Frith, 2006). In the context of reading this would suggest that
individuals with ASD may not integrate information within and between sentences. Evidence
for this lack of contextual/global processing during reading has most notably been
demonstrated using the homograph task. The homograph task involves participants reading
sentences aloud that contain a heterophonic homograph (i.e., a word with one spelling, but
two meanings and pronunciations) and readers with ASD have been reported to identify and
pronounce the dominant meaning of the homograph, regardless of sentence context (Frith &
BENCHMARK EFFECTS
4
Snowling, 1983; Happe, 1997; Jolliffe & Baron-Cohen, 1999; Lopez & Leekam, 2003 c.f.,
Snowling & Frith, 1986). However, the validity of these studies has been criticised (e.g.,
Brock & Bzishili, 2013; Brock & Caruana, 2014) and when other paradigms are adopted the
findings of impaired integration during reading tasks is inconsistent, with some researchers
finding poorer performance for individuals with ASD (e.g., Jolliffe & Baron-Cohen, 1999;
2000), whereas others do not (e.g., Au-Yeung, Kaakinen, Liversedge & Benson, 2015; Hala,
Pexman & Glenwright., 2007). The Theory of Complex Information Processing (CIP;
Minshew & Goldstein, 1998; Minshew, Goldstein & Siegel, 1997; Minshew, Williams &
McFadden, 2008) proposes low level ‘simple’ processes to be intact in ASD, but ‘complex’
processes that require the integration of information or the use of top down knowledge, to be
impaired. The evidence in support of this theory in relation to reading stems from studies
showing that adults and children with ASD perform comparably to matched controls on
batteries of standardised reading assessments that are defined as ‘simple’ (i.e., tasks that are
rule based and can be completed upon the basis of information explicitly stated in the text
e.g., word identification and grammar). In contrast performance is poorer compared to
controls at reading tasks that are defined as ‘complex’ (i.e., tasks that require processing
beyond explicit text information e.g., inferencing and comprehension, Minshew et al., 1995;
Minshew & Goldstein, 1998; Williams, Goldstein & Minshew, 2006).
The majority of research examining reading in ASD has focused on accuracy and
reaction time measures of performance. Although these studies are informative in relation to
offline reading performance, they provide little insight into specific aspects of online
cognitive processing associated with reading impairments in ASD. In the present work, we
recorded participants’ eye movements as they read naturally, to gain an insight into the online
linguistic processing of written language in ASD. A literature search indicated that only three
peer-reviewed articles have adopted this approach when examining natural reading in ASD
BENCHMARK EFFECTS
5
(Au-Yeung et al., 2015; Caruana & Brock, 2014; Sansosti, Was, Rawson, & Remaklus,
2013). This is the case, even though eye movement recording is one of the most widespread
methods used to examine reading in a typical population (e.g. Rayner, 1998; 2009).
The first study that used eye movements to examine reading in ASD was conducted
by Sansosti et al., (2013) who partially replicated an experiment originally conducted by
Saldaña and Frith (2007), whereby a group of children and adolescents with ASD read
vignettes consisting of two sentences, designed to invoke a bridging inference. Each vignette
either evoked an inference that required social or spatial knowledge and was followed by a
general knowledge question that was either related (e.g. 1a) or unrelated (e.g. 1b) to the
evoked inference (e.g. for the examples below the rocks hurt the cowboys).
1a.
The Indians pushed the rocks off the cliff onto the cowboys.
The cowboys were badly injured
Can rocks be large?
1b.
The Indians pushed the cowboys off the cliff onto the rocks.
The cowboys were badly injured
Can rocks be large?
Sansosti et al. (2013) replicated Saldaña and Frith’s (2007) main finding that both
typically developed (TD) and ASD readers responded to questions related to an inference
more quickly than questions that were unrelated to the inference. In addition, Sansosti et al.
(2013) reported that whilst reading the vignettes, the ASD participants had longer average
BENCHMARK EFFECTS
6
reading times, longer average fixation durations and made more fixations and regressions
overall, in comparison to the TD group. This finding highlights the discrepancy between
offline and online measures of reading. The lack of difference between the groups in response
times to the general knowledge questions indicates that the ASD group made bridging
inferences online. However, the eye movement data indicate that the online processing of the
vignettes was significantly more effortful for ASD participants. Sansosti et al. (2013)
interpreted their findings to be consistent with the WCC Theory and suggested that ASD
readers access world knowledge online, but have difficulty in the integration of this
knowledge into the discourse model. Although possible, this interpretation should be
considered tentative, as it is based upon global eye movement measures that are averaged
across the reading of the entire vignette. In order to make precise deductions about the exact
processes that may differ during reading, both global and localised eye movement measures
on critical, experimentally manipulated words and regions of a text are required. Therefore,
although this experiment gives evidence for the processing of text to be less efficient in ASD
readers, the time course of such differences were not fully explored. Consequently, the
aspects of cognitive processing that underpin eye movement differences are unclear.
The second study that used eye tracking to gain an insight into linguistic processing in
ASD examined two opposing hypotheses as to why ASD readers have previously been found
to perform poorly at the homograph task (Caruana & Brock, 2014). The homograph task
involves participants reading sentences aloud that contain a heterophonic homograph (i.e., a
word with one spelling, but two meanings and pronunciations). Readers with ASD are often
reported to identify and pronounce the dominant meaning of heterophonic homographs,
regardless of sentence context (Happé, 1997; Jolliffe & Baron-Cohen, 1999; Lopez &
Leekam, 2003; Snowling & Frith, 1983). Firstly, Caruana and Brock (2014) examined the
predominant conclusion drawn from these studies on the basis of the WCC Theory, that an
BENCHMARK EFFECTS
7
impairment in contextual processing is present in ASD. To test this hypothesis, fixation times
of students with various levels of self-reported ASD traits were measured (using the Autism
Quotient [AQ]; Baron-Cohen, Wheelwright, Skinner, Martin & Clubley, 2001), on words that
were highly predictable by sentence context (e.g. 2a, target word italicised) and words that
were unpredictable (e.g. 2b).
2a. Crocodiles live in muddy swamps most of the time.
2b. The girl knows about the swamps in the bush.
For TD readers, a highly constrained context is found to facilitate lexical access, with
predictable words being identified more quickly and fixated for less time, than unpredictable
words (e.g. Ehrlich & Rayner, 1981). Caruana and Brock (2014) found students with both
high and low AQ scores to show equivalent contextual facilitation, with shorter first fixation
durations on the target words when the sentence context was highly constrained (e.g. 2a), in
comparison to when it was not (e.g. 2b). This suggests that readers with high AQ scores or
‘subclinical levels of ASD’ process contextual information at the sentence level, similarly to
those with low AQ scores. This contributes to the developing literature that suggests the
contextual processing of language to be intact in ASD (e.g. Brock, Norbury, Einav, & Nation,
2008; Hahn, Snedeker & Rabagliati, 2015; Hala, et al., 2007; Henderson, Clarke, &
Snowling, 2011; Norbury, 2005) and indicates that the previous reports of poor performance
for ASD readers at the homograph task may not be a result of atypical contextual processing
(see Brock & Caruana, 2014; Brock & Bzishvili, 2013 for a discussion about methodological
issues that may have contributed to previous findings of poor performance in the homograph
task).
Caruana and Brock (2014) also examined an alternative hypothesis, that poor
BENCHMARK EFFECTS
8
performance on the homograph task may be a result of less efficient comprehension
monitoring. In other words, that individual’s with ASD do not track comprehension as
efficiently as TD readers and therefore may not detect when their interpretation of a sentence
is no longer coherent. To examine this hypothesis, participants read sentences that contained
a homograph (e.g. crane in 3a below) that was preceded by a determiner and therefore not
contextually constrained.
3a. The crane was slowly flying over the lake.
3b. The bird was slowly flying over the lake.
An offline cloze task indicated that for such sentences, participants identified the
dominant meaning of the word (e.g. crane – machine), as apposed to the subordinate meaning
(e.g. crane – bird). Following the homograph, the sentences contained a disambiguating verb
that indicated the dominant meaning of the homograph was incorrect and in fact that the
subordinate meaning of the homograph was intended. Typical readers were expected to fixate
the disambiguating verb for longer periods when it was preceded by a homograph in
comparison to a control synonym (e.g. 3b), as a result of the detection of an initial
misidentification of the ambiguous word. This manipulation however did not result in any
differences in fixation measures on the disambiguating verb, regardless of AQ score. An
interaction was found however, between heterophonic homographs (two pronunciations) and
AQ score, but not homophonic homographs (one pronunciation), with participants who had
high AQ scores having longer gaze durations (the sum of fixations from when a word is first
fixated until the reader leaves that word to either the left or right) on the target, in comparison
to those with low AQ scores. This may suggest that participants with high levels of selfreported autistic traits detected the discrepancy for heterophonic homographs, but that those
BENCHMARK EFFECTS
9
with low levels did not. This is in contrast to what was predicted, as it indicates that readers
with higher AQ scores were more sensitive to alternations of sentence meaning and therefore
were monitoring their comprehension more carefully. Note though, this finding only occurred
for a small number of heterophonic homograph stimuli and should therefore be treated with
caution. Therefore, Caruana and Brock (2014) did not find support for either hypothesis;
contextual processing appeared to be comparable across participants and the findings in
relation to comprehension monitoring were unclear. An additional more general observation
noted by Caruana and Brock (2014) was that a high AQ score was associated with longer
fixations for all sentence and condition types (although only reliable in the predictability
experiment). Note that this is similar to what Sanasosti et al. (2013) reported for their sample
of clinically diagnosed participants. If it is accepted that findings from a typical population
with high levels of self reported autistic traits can be generalised to individuals with a clinical
diagnosis, these findings suggest that contextual processing and comprehension monitoring at
the single sentence level are intact in ASD.
The third study was conducted by Au-Yeung et al. (2015), who had adult participants
with and without ASD read passages of text that contained either a sincere or ironic
statement. Previous work that has examined the processing of written irony in a typical
population has found ironic statements to require longer processing time and in turn longer
fixation times, in comparison to non-ironic statements (Filik & Moxey, 2010). It was
predicted that if readers with ASD are less sensitive to contextual information, as is predicted
by the WCC Theory, that the processing of ironic versus sincere utterances would not differ.
Alternatively, it was also predicted upon the basis of the CIP Theory, that the comprehension
of figurative language is more complex than the comprehension of literal language and
therefore readers with ASD should show increased processing disruption when encountering
ironic statements, as apposed to sincere statements. Surprisingly however, no differences in
BENCHMARK EFFECTS
10
first pass reading times (the time from when a region of text was initially fixated until that
region was left to the left or right) for ironic statements were found between TD and ASD
groups, with both displaying longer times for ironic than sincere statements, replicating what
has previously been found for a typical population. This suggests that the comprehension of
irony was as effortful for TD and ASD readers and is in contrast to both the WCC and CIP
Theory. The only group difference detected was that the ASD readers had longer total times
(the total amount of time spent fixating a region) for the critical regions in comparison to the
TD group. Note that no difference in first pass times were detected and that the increased
time that was observed was found for both ironic and sincere texts. The increased total times
appeared to be a result of the ASD participants re-reading the passages following initial
processing. The authors concluded that this difference was either a result of the ASD readers
requiring longer to develop a discourse representation of the text, or that they required longer
to conclude that their interpretation of the text was reasonable.
What should be evident from the above studies is that although very few have
examined online reading processes in ASD, those that have, focus upon aspects of semantic
and discourse processing that are predicted to be impaired by cognitive theories of ASD (e.g.
Frith & Happé, 1994; Minshew & Goldstein, 1998). However, only two of these studies
assessed a clinically diagnosed sample (Au-Yeung et al., 2015; Sansosti et al., 2009) and only
one of these demonstrated a difference in the initial extraction of information from the text
(Sansosti et al., 2009). Unfortunately this was evident in global reading time measures, which
makes the precise cause or timing of this effect difficult to determine. Therefore, the question
remains as to how the presence of this developmental disorder impacts upon the online
processing of written language.
The aim of the present work was to extend the emerging literature examining online
linguistic processing in ASD and to identify whether differences in fundamental lexical,
BENCHMARK EFFECTS
11
syntactic and semantic components of sentence processing are present in individuals with
ASD during natural reading. In order to achieve this we adopted robust, benchmark linguistic
manipulations from the eye movement and reading literature (Rayner, 1998; 2009).
Experiment 1. Lexical identification in ASD
Lexical identification refers to the processes a reader engages in to identify a word.
There are multiple computational models that specify the exact mechanisms involved in
visual word recognition in a typical population and each differ regarding particular aspects of
lexical processing (e.g. Coltheart, Rastle, Perry, Langdon & Ziegler, 2001; Grainger &
Jacobs, 1996; Rumelhart & McClelland, 1982). For example, the Dual Route Cascaded
Model (Coltheart et al., 2001) includes separate lexicons for phonological and orthographic
information and the Multiple Read Out Model includes variable response criteria that can be
altered dependent upon task requirements (Grainger & Jacobs, 1996). However, these models
are similar at a more basic level, in that they propose words to be identified by a matching
process between encoded orthographic information relative to a stored word representation in
the mental lexicon (cf. Siedenberg & McClelland, 1986). Regardless of the precise
mechanistic account underpinning lexical identification, in the current experiment, we were
interested in whether the time course of such lexical processing is similar in ASD readers.
Performance accuracy for word identification tasks varies between individuals with
ASD (e.g. Nation et al., 2006), however, for individuals without language impairment or
learning difficulties, performance on word identification tasks is generally found to be intact.
For example, children and adolescents with ASD have been found to use both phonological
and orthographic decoding strategies when identifying words, be as accurate as TD peers in
the reading of words aloud and have intact word comprehension (e.g. Heumer & Mann, 2010;
Minshew et al., 1995; Saldaña, et al., 2009).
There are also reports however, of atypical lexical processing in ASD that may be
BENCHMARK EFFECTS
12
suggestive of a difference in the timecourse of lexical identification. For example, Kamio,
Robins, Kelley, Swainson and Fein (2007) found children, adolescents and young adults with
ASD to respond as quickly and accurately as TD participants during a lexical decision task,
but to lack facilitaion from closley related semantic primes. Note that this is in contrast to
findings of normal semantic priming in ASD participants when other paradigms are adopted
(e.g. Hala et al., 2007; Lopez & Leekam, 2003; Kamio & Toichi, 2000; Toichi & Kamio,
2001), but this finding may be indicative of a slow in the access to word meaning. Further
support for less efficient lexical processing comes from Sansosti et al. (2009) and Caruana
and Brock’s (2014) experiments that were previously discussed above and have reported
participants with ASD to have longer average fixation durations in comparison to TD readers.
In this experiment, we aimed to directly examine the time course of online lexical
identification in adults with ASD in order to identify whether the efficiency of such
processing is comparable between ASD and TD readers. This is an important question to
address, because if online lexical processing is less efficient, this may cascade forward
impacting on later aspects of linguistic processing. Thus, prior to examining online higher
order linguistic processing, it is necessary to have knowledge of the proficiency of online
visual word identification.
We recorded the eye movements of TD and ASD participants as they read sentences
that included words manipulated to be either high or low in frequency. Word frequency is one
of the most reliable lexical characteristics that affect the speed of lexical identification. Word
frequency is a measure of how often a word occurs in the written language. High frequency
words (e.g. people) are identified more quickly than low frequency words (e.g. zombie)
because they have been previously encountered more often. This difference in identification
speed is reflected in fixation durations, with low frequency words being fixated for
significantly longer than high frequency words (e.g. Inhoff & Rayner, 1986; Rayner & Duffy,
BENCHMARK EFFECTS
13
1986). Connectionist models posit that each time a word is identified, the baseline level of
activity for that word increases and consequently, the more quickly a word reaches the
activity threshold necessary for identification (e.g. Rumelhart & McClelland, 1982).
Therefore, the frequency effect is thought to reflect a real difference in the time it takes to
identify a stored word representation on the basis of visually encoded orthographic
information. If lexical processing were less efficient in ASD, we would expect these readers
to have longer fixations on target words and possibly even show an increased magnitude of
the frequency effect, in comparison to the TD participants.
Method
Participants
A group of 19 adults with a formal diagnosis of ASD took part in the experiment, 18
of which had a diagnosis of Asperger’s disorder and 1 with high functioning ASD (3 females,
aged 18-52 years). Each ASD participant was administered module 4 of ADOS-2 (Lord et al.,
2012) and all but four met the autism spectrum cut off criteria. When these participants were
excluded the pattern of effects did not change and therefore all participants are included in all
analyses reported below. The control group consisted of 18 TD adults (4 females, aged 2052). Participants had normal or corrected to normal vision, were native English speakers and
did not differ in age t (35) = 0.94, p = .354 (TD M = 28 years SD = 9, ASD M = 31 years SD
= 10). The ASD group had a significantly higher number of self-reported autistic traits in
comparison to the TD group, as measured by the AQ (Baron-Cohen, et al., 2001) t (32) =
9.24, p < .001 (TD M = 15 SD = 8, ASD M = 37 SD = 6), but did not differ in verbal IQ t (35)
= 0.58, p = .621 (TD M = 117 SD = 11, ASD M = 118 SD = 11), performance IQ t (34) =
0.99, p = .331 (TD M = 112 SD = 11, ASD M = 116 SD = 15) or full scale IQ t (35) = 0.87, p
= .389 (TD M = 116 SD = 11, ASD M = 119 SD = 12), as measured by the Weschler
Abbreviated Scale of Intelligence (Wechsler, 1999). In addition, the two groups did not differ
BENCHMARK EFFECTS
14
in expressive language ability t (35) = 0.53, p = .599 (TD M = 88 SD = 6, ASD M = 87 SD =
6), as measured by raw scores on the sentence repetition subscale of the Clinical Evaluation
of Language Fundamentals II (Semel, Wiig & Secord, 2003). General reading ability was
assessed using raw scores from the York Assessment of Reading Comprehension (Snowling
et al., 2010) and performance between groups did not differ for a single word reading task; t
(33) = 0.51, p = .614 (TD M = 68 SD = 2, ASD M = 68 SD = 3) nor for a passage
comprehension task; t (34) = 0.35, p = .727 (TD M = 9.19 SD = 1.77, ASD M = 8.97 SD =
2.03). Participants were paid for their time and either visited the University of Southampton
to be tested, or were visited at their homes and completed the experiment in the psychology
departments mobile research unit.
Materials
Thirty-four sentence pairs were developed to include a target word located at
approximately the centre of each sentence that was either high (HF e.g. 4a below) or low
frequency (LF e.g. 4b). In the examples below the slashes denote region of interest
boundaries.
4a. |John walked to the large| office| yesterday morning.|
4b. |John walked to the large| cavern| yesterday morning.|
All target words were six letter nouns and significantly differed in frequency t (33) = 12.95, p
< .001 (HF M = 151.43 SD = 3.04, LF M = 3.04 SD = 10.04), but not in the number of
orthographic neighbours; t (33) = 0.45, p = .659 (HF M = 1.71 SD = 2.38, LF M = 1.44 SD =
2.67), mean bigram frequency; t (33) = 1.34, p = .190 (HF M = 3896.04 SD = 7614.47, LF M
= 3419.44 SD = 1353.73), number of syllables; t (33) = 0.37, p = .768 (HF M = 1.85 SD =
0.56, LF M = 1.88 SD = 0.48) or number of morphemes; t (33) = 0.81, p = .422 (HF M = 1.21
BENCHMARK EFFECTS
15
SD = 0.41, LF M = 1.26 SD = 0.45), as obtained from the SUBTLEX database (Brysbaert &
New, 2009). Target words were equally unpredictable t (33) = 1.66, p = .107 (HF M = .01 SD
= .03, LF M = .00 SD = .00), as rated by 12 undergraduate students on a cloze task. In
addition, sentences in each condition did not differ in plausibility t (33) = 1.65, p = .109 (HF
M = 4.35 SD = 0.49, LF M = 4.15 SD = 0.56) as rated by 13 undergraduate students (who had
not taken part in the cloze task) on a five point likert scale as to how likely it was that the
events they describe would occur (1 = very unlikely, 3 = quite likely, 5 = very likely). A full
set of the materials can be obtained by contacting the first author.
Procedure
Participants read sentences presented on a 19-inch LCD computer monitor (75 Hz) as
their head position was stabilised using a forehead and chin rest. The right eye was monitored
(viewing was binocular) using an Eyelink 1000 (SR Research) operating at a sampling rate of
1000 Hz. Before the experiment began, a calibration procedure was completed whereby
participants fixated three dots on the screen that appeared sequentially on a horizontal line
where the sentences were set to appear. Following calibration, a validation procedure was
completed to assure fixations were within 0.5° of each point.
At the start of each trial, participants fixated a dot on the far left of the screen, where
the first letter of each sentence was set to appear. If fixation was off-centre, participants were
re-calibrated. If calibration was accurate, the experimenter triggered a sentence to appear.
Sentences were presented one at a time and participants were instructed to read normally and
to press a button on a controller once they had finished reading each sentence. Sentences
were presented in random order. Following 50% of the sentences, participants were asked a
simple comprehension question about what they had just read. Participants responded with a
“Yes/No” answer using a button controller. Instructions as to which button corresponded to
BENCHMARK EFFECTS
16
each answer were included beneath each question. In total, the eye-tracking task took
approximately 25 minutes.
Design
A 2 X 2 design was employed, with sentence type as a within participants variable
(Experiment 1: High vs. Low frequency; Experiment 2: High vs. Low attachment) and group
(TD vs. ASD) as a between participants variable. The experimental sentences from
Experiment 1 and Experiment 2 (44 sentence pairs manipulated to include an ambiguous
prepositional phrase) were presented to participants within the same testing session. All
experimental sentences were divided into four separate lists that each contained only one
version of each sentence pair. In total each list consisted of 88 sentences; 34 that included a
frequency manipulation (17 HF, 17 LF), 44 manipulated to include an ambiguous
prepositional phrase (22 high attached, 22 low attached) and 10 practice sentences that were
presented prior to experimental sentences. Each participant read one of the four sentence lists.
Experiment 1: Results
Data preparation and analysis
Fixations below 80ms and above 800ms were removed from analysis, resulting in a
loss of less than 1% of the original fixation data. Trials when a participant blinked whilst
fixating the target word and when the trial was disrupted in some way were also removed,
resulting in a loss of 7.95% of data. In addition, data points that were more than 2.5 standard
deviations away from the mean (computed individually for each participant per condition)
were further excluded. This led to a loss of less than 3% of data from each fixation measure
calculated. Comprehension accuracy was high for both groups (TD M = 0.94, SD = .07; ASD
M = 0.97, SD = 0.04), which indicates that offline comprehension was not impaired.
For fixation measures, data was log transformed and confirmatory linear mixed
effects models (Baayen, Davidson & Bates, 2008) were computed using the lme4 package
BENCHMARK EFFECTS
17
(Bates, Maechler, Bolker & Walker, 2014) for R (R Core Team, 2015), with group (TD vs.
ASD) and frequency (HF vs. LF) defined as fixed categorical factors. Contrasts to obtain
main effects and the associated interaction were coded using the contr.sdif function from the
MASS library (Venables & Ripley, 2002). The full random structure was included (Barr,
Levy, Scheepers & Tily, 2013); with crossed random effects specified for both participants
and items. At the participant level, random slopes were included for frequency. At the item
level, random slopes were included for frequency, group and the associated interaction
between these factors. This resulted in the following syntax; lme(depvar ~ group*frequency +
(1 + frequency|participants) + (1 + group*frequency|items), data = data). Effects were
identified as significant if t > 2.
For binary variables (skipping and regressions), logistic linear effects models were
computed. For logistic models, when the full random structure was included (as specified for
continuous measures), models would not converge. Models were systematically trimmed of
parameters, beginning with the interaction in the random structure, until model convergence
was achieved. This resulted in a random structure whereby random slopes were only included
for frequency at the participant level, as depicted in the following syntax; glmer(depvar ~
group*frequency + (1 + frequency|participants) + (1 |items), data = data, family = binomial).
Binary variables were identified as significant when z > 2.
Global measures
To examine whether there were any basic sampling differences between the two
groups, the mean fixation duration, mean fixation count and total sentence reading time was
calculated across each trial. Model parameters and observed means for each of these
measures are presented in Table 1.
No difference between groups was found for mean fixation duration, but an effect of
frequency was found, with sentences that included a low frequency word receiving longer
BENCHMARK EFFECTS
18
average fixation durations than those that included a high frequency word. No interaction
between group and frequency was detected. For mean fixation count, a reliable difference
between groups was found, with ASD readers making more fixations over the course of each
trial, in comparison to TD readers. However, the number of fixations participants made over
the course of each trial was not affected by the frequency of the target word, and there was no
interaction. Total sentence reading times were also found to be significantly longer for the
ASD group, in comparison to the TD group, but there was no effect of frequency and no
group by frequency interaction.
These global measures indicate that the ASD group extracted information during
fixations at a similar speed to TD readers, but made more fixations during the course of each
trial and also had longer sentence reading times, in comparison to TD participants.
**INSERT TABLE 1 ABOUT HERE**
Target analysis
A region of interest was created around the target noun. For the target word, the
following eye movement measures were calculated; skipping, first fixation duration (the
duration of the first fixation on the target), single fixation duration (the duration of a fixation
on the target, when this was the only fixation made on this word during first pass reading),
gaze duration and total time.
All reading measure means and standard deviations for each region of analysis are
included in Table 2 and the model parameters are presented in Table 3. Word frequency did
not have an affect on the probability of the target word being skipped. However, frequency
was found to affect the duration of first fixations, single fixations, gaze durations and total
times on the target word, with each of these measures being greater for low frequency words
in comparison to high frequency words. No interactions were found for any measure and the
BENCHMARK EFFECTS
19
only group difference detected was that the readers with ASD had longer total times for the
target region, in comparison to TD readers.
Based on this analysis, there is no evidence for the hypotheses that lexical processing
is less efficient in ASD. However, in order to infer the extent to which our data reflect a null
effect of group, as apposed to a Type II error, we computed Bayes factor (Kass & Raftery,
1995; Rouder, Morey, Speckman & Province, 2012) for the first fixation LME model on the
target word reported above (with a frequency by group interaction), when compared to a
denominator model that had the same random structure, but only included frequency as a
fixed effect. Bayes factor.is a form of Bayesian analysis whereby one can quantify the
relative evidence (probability) for apposing hypotheses, based on the change of prior odds to
posterior odds, as a result of the observed data. One of the benefits of this approach is that
one can assess evidence in favour of a null hypothesis, which is not possible through more
traditional null hypothesis significance tests. A detailed description of this analysis is beyond
the scope of this paper, however interested readers are referred to (Kass & Raftery, 1995;
Rouder et al., 2012; Wagenmakers, 2007). Thus, to be clear, we directly compared our
original model to one that did not specify group as a predictor. We chose to focus this
analysis on the first fixation duration data because this measure is highly influenced by the
lexical properties of a word and this measure for this region was the first point that frequency
had an effect on fixations for the TD group. A Bayes factor of less than 1 would favour the
denominator model and a value above 1 would favour the original model (that included a
group by frequency interaction). We computed the Bayes factor in R (R Core Team, 2015)
using the BayesFactor package (Morey & Rouder, 2013) with 100,000 Monte Carlo iterations
and with g-priors scaled to r = 0.5 for fixed effects. The Bayes factor for the original model,
when compared to the denominator model was 0.045. Based on Jeffrey’s (1961) evidence
categories for Bayes factor, this provides strong evidence in favour of the denominator model
BENCHMARK EFFECTS
20
that did not include group as a predictor. Moreover, the denominator model is 22 times more
likely, based on our data, than the original model. This supports the conclusion that the ASD
and TD groups did not appear to differ in the efficiency with which they lexically processed
the target words.
Note that the lack of group differences in the early stages of target word processing
indicates that the increased number of fixations and increased time spent reading the
sentences in the global analyses, is not a result of differences in the early stages of lexical
processing. In order to determine the possible cause of these global differences, the spatial
and temporal characteristics of the extra fixations ASD participants made are examined in the
analysis below.
First pass reading
In the following analyses, the start and end regions of the sentences were examined, in
addition to the target word. The start region included all words prior to the target. The end
region included all the words following the target. It is possible that the increased sentence
reading times and number of fixations made by the ASD group are a result of longer and
more effortful first pass reading. To examine whether such a difference exists, gaze durations
for the start and end region, and the proportion of regressions made out of the target and end
region during first pass was examined.
Recall that no group differences were present for gaze durations in the target region.
Similarly, for both the start and the end region, no effect of group, condition or an interaction
was detected. Furthermore, for both the target and end region, there were no differences
between groups or frequency in the proportion of regressions made out of these regions
during first pass of the sentences. What this suggests is that the initial processing of each
region of the sentence did not differ between groups.
Second pass reading
BENCHMARK EFFECTS
21
Given the lack of difference detected between groups during the initial processing of
the sentences, it is possible that the increased reading and fixations made by ASD participants
occurred during second pass reading, that is, after the participants had read each sentence
once in its entirety.
Total second pass reading times were first calculated across the entire sentence
(summed total times – summed gaze durations, see Table 1). Readers with ASD were found
to have larger second pass reading times for the entire sentence, in comparison to TD readers,
but there was no effect of condition or any interaction. To determine whether there was a
particular area of sentences that the ASD group were re-reading, second pass reading times
were computed for each region individually (total time- gaze duration) and the ASD group
were found to have longer second pass reading times for the start and end region of the
sentence. However, no effect of frequency or any interactions were detected.
To examine whether the ASD group were making more regressions into either the
start or target region, in order to engage in second pass reading, the proportion of trials a
regression was made into these regions was examined. No effect of group or word frequency
was found for either region. This suggests that there was no reliable difference in the
proportion of trials with which the two groups regressed into the start and target regions, but
that when the ASD group did make such a regression, they spent longer re-reading each
region of the sentences, compared to the TD group.
**INSERT TABLE 2 & 3 ABOUT HERE**
Experiment 1: Discussion
We examined online lexical processing in ASD by measuring participant’s eye
movements as they read sentences that contained a target word manipulated to be of high or
low frequency. The target analyses revealed that both TD and ASD readers showed normal
word frequency effects for all target fixation measures. Fixations were significantly longer on
BENCHMARK EFFECTS
22
low frequency words, in comparison to high frequency words. This finding of a normal
frequency effect extends our current knowledge in relation to lexical processing in ASD, as it
demonstrates that in addition to intact performance accuracy for isolated word identification
tasks (e.g. Frith & Snowling, 1983; Mayes & Calhoun, 2006; Minshew et al., 1995; Saldaña
et al., 2009), that the processes engaged in to identify a word during normal reading appear to
be comparable between ASD and TD readers. This is in line with cognitive theories of ASD
that suggest low level ‘bottom up’ processing to be intact (e.g. Minshew & Goldstein, 1998).
Our findings are inconsistent however with Sansosti et al.’s (2009) study that reported
ASD readers to have longer average fixation durations. We found no evidence of such a
difference in our data. This inconsistency may be attributable to the differences between the
stimuli employed by our own and Sansosti et al.’s (2013) experiment. The vignettes in
Sansosti et al.’s (2013) study were designed to evoke a bridging inference, whereas our
sentences required no inferences to be made in order to comprehend sentence meaning. Our
finding therefore gives indirect support for Sansosti et al.’s (2013) interpretation that the
larger average fixation durations observed for ASD participants, reflected more effortful
processing in relation to the computation of an inference, as apposed to differences in lexical
processing.
In the global analyses, we found participants with ASD to spend longer reading
sentences and make more fixations overall, during the course of a trial. This is consistent with
what Sansosti et al. (2009) reported, however the time course of the increased reading time
found for their experiment was not reported, making the cause of these global effects unclear.
In our analysis we examined the time course of the increased reading times for the ASD
participants, in order to determine whether these differences were related to our experimental
manipulation, and the nature of such increased reading. This analyses revealed no difference
between the groups for first pass reading times or first pass regressions, which suggests that
BENCHMARK EFFECTS
23
the speed and manner in which an initial sentence interpretation was constructed to be alike
for both TD and ASD participants. Crucially, although our experiments were not designed to
examine integrative processes, this finding is inconsistent with theories that suggest such
processes to be atypical in ASD (e.g. Frith & Happé, 1994, Minshew & Goldstein, 1998). If
integration were more effortful for ASD readers, we would have found longer gaze durations
in each region and a larger proportion of first pass regressions being made. This was not the
case.
The analyses also suggested that the increased sentence reading times for ASD
participants appeared to be wholly a result of the these participants re-reading the sentences
for significantly longer periods of time than the TD group. Note that this increased re-reading
appeared to be unrelated to our target word manipulation, with re-reading occurring equally
often for sentences that contained low and high frequency words and is consistent with what
Au-Yeung et al. (2015) report for ironic and sincere texts. In addition, no group differences in
the proportion of regressions made between groups were found. This is inconsistent with
Sansosti et al.’s (2009) finding of increased regressions being made by ASD participants and
indicates that our ASD group were as likely as the TD group to make a regression out of a
region and also that the target of this regression did not differ. What did differ, was that once
having made a regression, the ASD readers spent longer re-reading the sentences.
The cause of this re-reading behaviour is unclear, but it seems unlikely to be a result
of a linguistic processing deficit per se, as any difficulty associated with the extraction of
sentence or word meaning would have been evident during first pass reading. We speculate
that this difference may reflect a ‘checking’ strategy adopted by the ASD group. Recall that
there were comprehension questions after 50% of sentences and this may have contributed to
ASD participants being more cautious of their sentence interpretation and consequently
spending longer than TD participants re-reading sentences to ensure that they could answer
BENCHMARK EFFECTS
24
the questions accurately. What is important is that the increased second pass reading was not
a result of difficulties in the initial extraction of sentence meaning.
Experiment 2: Syntactic and Semantic Processing
It is common to encounter structural ambiguities within natural language; however,
these often go unnoticed due to the parsing preferences readers hold. For example, in the
sentence below (5) the prepositional phrase is syntactically ambiguous.
5. Jane hit the man with the handlebar moustache.
A reader can either attach the prepositional phrase high as a modifier to the verb hit,
or low as a modifier to the noun phrase the man. The two possible syntactic structures result
in a different sentence interpretation, with the first suggesting that Jane hit the man using a
handlebar moustache and the latter suggesting that Jane hit the man who had a handlebar
moustache. For sentence 5, if a high attached structure is adopted, the sentence meaning is
implausible and reading is disrupted (e.g. Rayner, Carlson & Frazier, 1983; Taraban &
McClelland, 1988). This is because the reader has to re-evaluate their initial parse of the
sentence, to adopt the alternative, low attached structure. This disruption is evident in the eye
movement record as increased fixation durations on the disambiguating noun and increased
regressions out of this region, to re-read previous parts of the sentence (e.g. Joseph &
Liversedge, 2013; Rayner et al., 1983).
There are several theoretical positions adopted as to why certain structures are
preferred and initially adopted by readers. The Garden Path theory (Frazier & Rayner, 1982)
posits that sentences are parsed according to two principles; Minimal Attachment and Late
Closure. Minimal Attachment is a rule according to which a reader will always initially build
the simplest syntactic structure, and the Late Closure rule stipulates that the parser will
always attach a phrase to the currently open phrase structure. An alternative theoretical
BENCHMARK EFFECTS
25
position is that of Constraint-Based models where the language processor is assumed to be
interactive and higher order information such as contextual information can influence initial
decisions (e.g. MacDonald, Pearlmutter & Seidenberg, 1994). There are other theoretical
accounts of parsing too (e.g. Construal, Frazier & Clifton, 1996; Race Based Parsing, Van
Gompel, Pickering & Traxler, 2000; Good Enough Parsing, Ferreira, Bailey & Ferraro,
2002). For present purposes, we will again sidestep the issue of which theoretical account
provides the most adequate account of processing, and instead simply accept that processing
biases exist for sentences with particular linguistic characteristics. On this basis, in
Experiment 2, analogous to Experiment 1, we will investigate how comparable on-line
syntactic processing is in TD and ASD readers.
The research that has examined syntactic processing in ASD during spoken language
processing has reported mixed results (for a review see Eigsti, Marchena, Schuh & Kelley,
2011), and little work has been conducted to examine the syntactic processing of written
sentences. In a study assessing the impact of ASD and language phenotype on reading
comprehension, Lucas and Norbury (2014) had children with ASD and typical language
skills (ALN); children with ASD and language impairments (ALI) and TD participants read
sentences that were syntactically ‘scrambled’. All groups showed equivalent disruption to
reading times when reading scrambled sentences, however there was a trend to suggest that
the ALI participants took longer to read syntactically legal sentences, in comparison to the
TD and ALN participants, who did not differ. This suggests that syntactic processing in ASD
may be intact, but that the presence of language impairment has an impact upon the
efficiency of syntactic processing during reading (Lucas & Norbury, 2014). Stockbridge,
Happé and White (2014) found that children with ASD were as accurate as TD children in the
identification of the indirect object of a verb in sentences that varied in structural form. In
addition, both TD children and children with ASD showed higher accuracy for sentences
BENCHMARK EFFECTS
26
containing a preposition phrase construction (e.g. Toby read the book to Jenny) in
comparison to a double object construction (e.g. Toby read Jenny the book), indicating that
both TD and ASD children benefited from the indirect object being more explicit. These
results are consistent with Lucas and Norbury’s (2014) finding and suggest that syntactic
processing of written sentences is intact in ASD. The important finding from these
experiments in relation to the current work is that the individuals with ASD appeared to
exhibit comparable syntactic processing to that observed in TD children.
There is evidence, however, of a delay in the detection of syntactic errors in ASD.
Koolen, Vissers, Hendriks, Egger and Verhoeven (2012) had adults with ASD take part in a
‘single’ or ‘dual’ level reading task where sentences were presented in a serial viewing
paradigm. For the dual task participants had to detect orthographic errors (letter substitutions
e.g. the dog berks/barks) and syntactic errors (verb agreement errors e.g. she takes the broom
and sweep/s the floor), and in the single task participants had to detect only one error type
(lexical or syntactic). Koolen et al. (2012) found that ASD and TD participants did not differ
in their performance accuracy for either task and that for the single task reaction times did not
differ. However, during the dual task, ASD participants were slower to detect both
orthographic and syntactic errors, in comparison to the TD group who only showed a slowed
response for orthographic errors. The lack of difference between groups for the single task
was taken as evidence of intact lexical and syntactic processing in ASD. The slow in the
detection of both types of errors under dual task conditions however suggested that when
both lexical and syntactic information had to be monitored simultaneously that processing
became less efficient. Furthermore, Koolen, Vissers, Egger and Verhoeven (2014) replicated
this study but also recorded event related potentials (ERP) and found that when an
orthographic error was encountered, both TD and ASD participants emitted comparable P600
(a positive waveform that has previously been associated with error detection) responses
BENCHMARK EFFECTS
27
under both task conditions. However, larger P600 amplitudes were found for ASD readers
upon the detection of syntactic errors under both dual and single task conditions. The authors
concluded that it is reduced attentional modulation, as apposed to atypical linguistic
processing that results in poor performance for reading tasks in ASD and that the increased
P600 amplitudes for syntactic errors reflect a difficulty in the processing of language during
complex tasks. It is therefore possible that more basic differences in the online processing of
information are present in ASD during normal reading conditions, which requires the
simultaneous processing of lexical and syntactic information.
From the work mentioned above, it would seem that individuals with ASD are able to
accurately complete tasks requiring syntactic processing, however it is unclear whether the
efficiency of such processing is similar between ASD and TD groups. A recent study (Riches,
Loucas, Baird, Charman & Simonoff, 2015) had adolescents with ASD complete an aural
comprehension task for sentences that contained an ambiguous prepositional phrase (e.g. the
girl approached the butterfly on the log). Response times were longer for each group when
the picture displayed reflected the least plausible interpretation, suggesting that both groups
had to reanalyse their initial plausible interpretation of the sentence, prior to making a
response. In addition, participants with ASD had longer response times overall, indicating
that they required longer to complete the task in all conditions. However, as noted by the
authors, it is unclear whether this increase in response time reflects differences in the
efficiency of language processing, or a more task oriented effect, such as increased scanning
of the pictures. In addition, there was a trend to suggest that ASD participants may hold a
stronger high attachment bias, in comparison to a TD group. However, again, the exact cause
of this effect is difficult to pinpoint given the analysis of reaction times. What this experiment
does elucidate however is that when presented with visual displays of an image, individuals
BENCHMARK EFFECTS
28
with ASD were able to reanalyse their initial interpretation of a sentence and select an
alternative.
In the present study we aimed to further examine the processing of such sentences in
ASD, however we were specifically concerned with such processing and its time course
during natural reading (as apposed to during listening, and in the absence of visual cues). We
had participants with and without ASD read sentences that contained an ambiguous
prepositional phrase and were designed to evoke a high attachment preference. We were
more specifically interested in whether the syntactic preferences held by ASD participants for
ambiguous sentences are similar to those of TD readers and also whether the time course of
disruption to reading by an initial syntactic misanalysis was comparable between groups. We
predicted that if readers with ASD did not hold a high attachment preference for sentence
stimuli, they would not show disruption to reading upon encountering the disambiguating
noun in low attached sentences. Alternatively, if readers with ASD do hold a high attachment
preference, but are less efficient in the recovery from an initial syntactic misanalysis, we
would expect to find increased disruption to reading for low attached sentences, in
comparison to the TD group.
The second aim of this experiment was to assess whether ASD readers use world
knowledge online during reading. The detection of an incorrect interpretation for low
attached sentences is dependent upon a reader’s ability to evaluate their sentential
interpretation against their knowledge of the world. Therefore, in Experiment 2 we were able
to directly assess the efficiency of online world knowledge use during reading.
Impairments in performance accuracy are often found for reading comprehension and
inferencing tasks in ASD (Heumer & Mann, 2010; Minshew et al., 1995; Jolliffe and BaronCohen, 1999) and for each of these tasks, the processing of world knowledge and then the
integration of this information into the discourse model is critical (Graesser, Singer, &
BENCHMARK EFFECTS
29
Trabasso, 1994; McKoon & Ratcliff, 1992). Furthermore, it has been suggested that top down
processing may be atypical in ASD (e.g. Minshew & Goldstein, 1998) and in the context of
reading, this may result in the efficiency with which world knowledge is known to be used
online being compromised.
Few studies have directly examined the use of world knowledge during reading in
ASD and those that have report mixed results. Saldaña and Frith (2009) had participants read
two sentence vignettes intended to produce an inference. Participants were then asked a
comprehension question that was or was not related to the inference the vignette was intended
to evoke. Reaction times to the questions did not differ between TD and ASD readers and
both groups showed faster responses when the question was related to the evoked inference.
This was interpreted to suggest that the use of world knowledge is intact in ASD, as ASD
participants must have incorporated such information into the discourse model in order to
demonstrate an inference priming effect on question reading times. Recall that Sansosti et al.
(2013) reported their sample of ASD readers to make longer average fixation durations, more
fixations and more regressions on average than their TD group when reading the vignettes
originally used in Saldaña and Frith’s (2009) experiment. It is therefore possible that subtle
differences in the online use of such knowledge are present in ASD and may impact upon the
comprehension of larger portions of text, but are undetectable via relatively coarse measures,
such as reaction times. Consistent with this hypothesis, Wahlberg and Magliano (2004) found
subtle differences between TD and ASD participants in the use of world knowledge to aid
recall of ambiguous texts. Participants read passages that were either preceded by a title
(informative or non-informative) and with, or without, a primer paragraph that explicitly
described the events the stories referred to. Overall, when a cue was present (title or primer
paragraph), TD readers recalled more clauses in total and recalled more clauses that
demonstrated world knowledge had been integrated into the discourse model, in comparison
BENCHMARK EFFECTS
30
to the ASD group. Note however that the number of clauses recalled that were inferred on the
basis of world knowledge did not differ between groups. This suggests that readers with ASD
accessed world knowledge online during reading, however they did not use this information
as efficiently as TD readers to assist in the disambiguation and recall of the ambiguous texts.
This is similar to the conclusions of Sansosti et al. (2013).
In Experiment 2, we aimed to directly examine the time course and efficiency of
world knowledge use during reading of syntactically ambiguous sentences in ASD. We
predict that if participants with ASD are less efficient in the use of world knowledge during
reading, then there will be a delay in their detection of an initial misanalysis of low attached
sentences, in comparison to TD readers.
Method
Participants, procedure and experimental design were identical to Experiment 1.
Materials
Forty-four sentence frames were devised that included a prepositional phrase that
could either be attached high (HA) to a verb (e.g. 6a) or low (LA) to the noun phrase that
immediately preceded the preposition (e.g. 6b). Both sentence types were identical aside from
the disambiguating target noun.
6a. |Charlie| demolished| the dilapidated house| with| a huge| crane| last| year.|
6b. |Charlie| demolished| the dilapidated house| with| a huge| fence| last| year.|
To confirm that the chosen verbs elicited a high attachment preference in a typical
population, 13 undergraduate students completed a cloze task and the experimenter went
through each sentence completion with each student, to clarify any possible ambiguities. The
students completed the sentences using a noun that resulted in a high-attached structure, 98%
BENCHMARK EFFECTS
31
of the time. On average, the target words did not differ in length t (43) = 0.17, p = .868 (HA
M = 5.59 SD = 1.69, LA M = 5.57 SD = 1.59), frequency t (43) = 0.64, p = .524 (HA M =
22.61 SD = 39.20, LA M = 27.83 SD = 61.08), number of orthographic neighbours t (43) =
1.25, p = .225 (HA M = 4.66 SD = 5.55, LA M = 5.89 SD = 6.53), mean bigram frequency t
(43) = 1.34, p = .186 (HA M = 3587.87 SD = 1395.70, LA M = 4232.46 SD = 2079.95),
number of morphemes t (43) = 0.38, p = .767 (HA M = 1.25 SD = 0.53, LA M = 1.23 SD =
0.48) or number of syllables t (43) = 0.83, p = .412 (HA M = 1.59 SD = 0.76, LA M = 1.52
SD = 0.73), as retrieved from the SUBTLEX database (Brysbaert & New, 2009). To assure
the high and low attached sentences did not differ in plausibility, 15 undergraduates (who had
not completed the cloze task) rated how likely it was that the events described in the
sentences would occur on a 5 point likert scale (1 = extremely unlikely, 3 = quite likely, 5 =
extremely likely). To avoid the low attachment of the prepositional phrase acting as a
confounding variable that may cause participants to rate these sentences as less likely, the
sentences rated were an unambiguous description of the events depicted in the sentences (e.g.
see 7a and 7b).
How likely is it…
7a. that a crane would be used to demolish a huge, dilapidated house.
7b. that a huge, dilapidated house that has a fence, would be demolished.
The order of items was pseudo-randomised so that the same sentence versions (e.g. 7a & b)
were at least 20 items apart. Participant ratings indicated that the events described in the
sentences did not differ in plausibility t (43) = 1.30, p = .200 (HA M = 3.39 SD = 0.78, LA M
= 3.27 SD = 0.79). A full set of materials can be obtained by contacting the first author.
Experiment 2: Results
BENCHMARK EFFECTS
32
Data preparation
Sentences were divided into eight regions, three of which were of primary interest; the
pre-target region which consisted of a determiner and adjective that immediately followed the
preposition; the target, which consisted of the disambiguating noun and the post target region
which included the one or two words that followed the target (e.g. 6a & b).
Data exclusion procedures where the same as Experiment 1, which resulted a loss of
4.74% of the original data, and a loss of less than 3% of data from each fixation measure
analysed. Sentence comprehension was high for both participant groups and therefore,
similarly to what we found for Experiment 1, any differences detected in the eye movement
data did not appear to impact upon offline comprehension outcomes (TD M = .96 SD = .04;
ASD M = .97 SD = .05). Analyses procedures were also identical to Experiment 1, but with
each model containing attachment condition (HA vs. LA) in place of frequency.
Global measures
Global measures were calculated across trials to identify whether there were any basic
sampling differences between TD and ASD readers during the reading of the syntactically
ambiguous sentences. Means, standard deviations and model parameters for the global
analysis are displayed in Table 4. No differences between groups or sentence types were
detected for mean fixation duration and there was no interaction. For mean fixation count,
there was a significant effect of attachment, with both groups making more fixations when
sentences were low attached, in comparison to high attached. In addition, there was a
significant effect of group, with ASD readers making more fixations overall in comparison to
TD readers, but there was no interaction. A similar pattern was also found for sentence
reading time, with both groups having longer reading times for low attached sentences, in
comparison to high attached sentences, and ASD readers having longer reading times overall,
BENCHMARK EFFECTS
33
but again there was no interaction. These findings replicate what was found for Experiment 1
and are explored in more detail later.
**INSERT TABLE 4 ABOUT HERE**
Pre-target region
Means, standard deviations and model parameters for all region analyses are
displayed in Tables 5, 6 and 7. Note that the pre-target region was identical across both high
and low attached sentences and as a result, no effect of attachment was found for skipping,
first fixation durations, single fixation durations, gaze durations, go past times (the time from
when a word is first fixated until the eyes leave this region to the right, including all time
spent re-reading previous regions of the sentence) or the proportion of regressions made out
of this region. In addition, group membership did not have an effect on any of these
measures, and there were no interactions. An effect of both attachment and group was found,
however, for total times, with the ASD group fixating this region for longer than the TD
group overall and both groups spending longer fixating this region when the sentence was
low attached. In addition, an effect of attachment for the proportion of regressions made into
the pre-target region was found, with both groups regressing back into the pre-target region
on a higher proportion of trials when the prepositional phrase was low attached, in
comparison to when it was high attached. No group difference or interaction was detected.
**INSERT TABLES 5, 6 AND 7 ABOUT HERE**
Target region
The target word disambiguated the prepositional phrase attachment and was therefore
the point at which participants were expected to first show disruption to reading for low
attached sentences. One unexpected finding was that ASD readers were less likely to skip the
target region in comparison to TD readers, but this did not differ between attachment
conditions and there was no interaction. No effect of group or attachment was found for first
BENCHMARK EFFECTS
34
fixations durations, indicating that during the very early stages of target processing, neither
TD nor ASD readers had detected an incorrect interpretation. An effect of attachment was
first present in single fixation durations, with both groups spending longer fixating target
words in low attached sentences, in comparison to high attached sentences. No group
differences or interactions were detected, indicating that the onset and severity of initial
processing disruption was equivalent for TD and ASD readers. Similarly, gaze durations and
go past times were found to be longer for target words within low attached sentences. Again,
no group differences or interactions were found for these measures. Expectedly, an effect of
attachment was also present in total times, with longer time being spent fixating target words
for low attached sentences in comparison to high attached sentences, but no group difference
or interaction was present. In addition, both groups made a higher proportion of regressive
fixations into the target region when the sentence was low attached, in comparison to high
attached, but the proportion of first pass regressions made out of the target region did not
differ between groups or attachment conditions.
The results above provide no evidence for the hypothesis that readers with ASD may
display increased disruption for low attached sentences, or a delay in the onset of disruption,
in comparison to TD readers. In order to examine the supportive evidence for this null effect,
we again compared the single fixation duration LME reported above to a denominator model
that only included prepositional phrase attachment as a fixed effect using Bayes factor. We
were, therefore, once again able to directly compare the relative evidence for the original
model to a model that excludes group as a predictor. We analysed the single fixation duration
data because this was the first point in time an effect of prepositional phrase attachment was
detected for the TD group, and this is a critical measure in relation to our hypotheses, where
differences in the magnitude or onset of the effect might be expected to occur. A Bayes factor
of 0.037 was computed using the same methods as detailed in Experiment 1. This indicates
BENCHMARK EFFECTS
35
that there is strong evidence that the denominator model is more probable and is 27 times
more likely based on the observed data, than the original model that included a group by
prepositional phrase attachment interaction. Our analysis provides evidence in favour of a
null effect of group in our data during the initial processing of the target word.
Post target region
The post target region was examined to assess the extent to which low attached
sentences continued to disrupt reading, following processing of the target noun. No spillover
effects were found, with no effect of group, attachment or any interactions for skipping, first
fixations, single fixations or gaze durations. Attachment did however influence go past times,
with both groups taking longer to progress to the right, past the post target region, when the
sentences were low attached, in comparison to high attached sentences. In addition, there was
a difference between groups, with ASD participants having longer go past times overall for
this region, but this did not interact with condition. For total times, there was a main effect of
attachment and group, with all participants spending longer fixating this region for low
attached sentences and the ASD group spending longer in this region overall. Participants
made significantly more regressions out of the post target region when the sentences were
low attached and this did not differ or interact with group. There were no differences between
groups or sentence types in the proportion of regressions made into this region.
First Pass Reading
To examine whether the increased sentence reading time and number of fixations
made by ASD participants in Experiment 2 was a consequence of increased first pass or
second pass reading times, additional first and second pass measures were examined. Recall
from the main analyses, there were no group differences or interactions in gaze durations for
the pre-target, target or post-target region. This was also true for the start, verb, noun,
preposition and end regions. Furthermore, there was no effect of attachment and no
BENCHMARK EFFECTS
36
interactions. The probability of readers making a regression out of each region during first
pass reading was also examined. If ASD readers made more regressions during first pass this
may have contributed to the increased sentence reading times. Recall that in the main
analysis, no group differences or interactions were detected in the pre-target, target or post
target regions. Similarly, there were no reliable differences between the proportion of first
pass regressions made between groups for any other region and there was also no effect of
attachment. These analyses suggest that there were no differences in the way the two groups
sampled information during first pass reading.
Second Pass Reading
Analysis of second pass reading time for the entire sentence (see Table 4) indicated
that the ASD group had longer second pass reading times overall, in comparison to the TD
group and that both groups engaged in more re-reading for low attached sentences in
comparison to high-attached sentences. To determine whether this increase in second pass
reading of the sentences was a result of the ASD group re-reading a particular area of the
sentence, second pass times were calculated for each region. The ASD group had longer
second pass times in comparison to the TD group for the start, noun, pre-target, target and
end. No differences between second pass reading of the preposition, post target or verb
regions were detected, but there is a numerical trend in the data, consistent with the above
findings. In addition, a reliable effect of attachment was detected for the noun and pre-target
region, indicating that when sentences were low attached, both groups re-read these areas of
the sentence more so, to assist in structural reanalysis. No interactions or effects of
attachment were detected for any other region. Recall from the main analysis that both groups
made a higher proportion of regressions into the target and pre-target regions, when the
sentences were low attached. This was also true for the noun and preposition, indicating that
this information was crucial for both groups, when reanalysing an initial structural
BENCHMARK EFFECTS
37
interpretation. These findings replicate the findings from Experiment 1, with ASD readers
spending significantly longer re-reading each region of the sentence, regardless of sentence
condition. Once again this demonstrates that this increased re-reading is independent of our
sentence manipulation.
Experiment 2: Discussion
We examined aspects of syntactic and semantic processing by having participants
read sentences that contained an ambiguous prepositional phrase. Our first research question
was related to whether readers with ASD hold similar syntactic preferences to TD readers.
Both TD and ASD readers displayed disruption to reading for low attached sentences upon
fixation of the target word, which indicates that similarly to TD readers, adult readers with
ASD hold a high attachment preference. Furthermore, the severity of this disruption did not
differ between groups, which suggests that readers with ASD are also as efficient as TD
readers, in the recovery from an initial structural misanalysis. These findings suggest that the
syntactic parsing of written sentences is intact in ASD.
Our results are consistent with Lucas and Norbury’s (2014) study that demonstrated
children with ASD to show similar disruption to TD children when reading syntactically
scrambled sentences. In addition, similarly to Riches et al.’s (2015) study we found adult
readers with ASD to be successful in the reanalysis of an initial interpretation of a sentence
containing an ambiguous prepositional phrase. Note however there was no evidence that
adults with ASD show a stronger high attachment preference, as a trend in Riches et al.’s
(2015) data suggested. Our findings are inconsistent with Koolen et al.’s (2012; 2014) work
that suggested syntactic processing to be less efficient in ASD. This inconsistency may be
attributable to the difference in task types adopted in our own and Koolen et al.’s (2012;
2014) experiments. The current study simply required participants to read sentences for
comprehension, whereas Koolen et al.’s (2012; 2014) studies required readers to detect and
BENCHMARK EFFECTS
38
respond to lexical and syntactic errors during a serial viewing paradigm. This requires
additional memory load and decisional processes that are not necessary during natural
reading.
Our second research question was related to the use of world knowledge online during
reading. We predicted that if such processing was less efficient that there would be a delay in
the onset of reading disruption for the target word for low attached sentences. However, no
such delay was detected, with both groups first showing disruption in single fixation
durations. What this suggests, is that counter to our hypothesis and to theories that predict top
down, higher order processing to be atypical (e.g. Minshew & Goldstein, 1998), ASD readers
had access to and made use of world knowledge online and did so as efficiently as TD
readers. This finding is consistent with Saldaña & Frith’s (2009) study that reported
participants with ASD to use world knowledge in order to make an inference, and in turn
respond more quickly to comprehension questions that were related to such an inference.
However, our finding is inconsistent with Wahlberg and Magliano’s (2006) report of ASD
readers being less efficient in the use of world knowledge. Note that Wahlberg and
Magliano’s (2006) study examined the use of such knowledge to aid in the recall of
ambiguous texts and it is therefore possible that the difference reported by Wahlberg and
Magliano (2006) is specifically related to memory functions following text comprehension,
as apposed to linguistic processing per se.
There were two subtle differences found between the participant groups in the early
processing of the sentences examined for Experiment 2. Firstly, the ASD readers were less
likely to skip the target word of both high and low attached sentences, in comparison to TD
readers. The lack of difference in skipping found in Experiment 1 and all other critical
regions of the sentences in Experiment 2, suggests that the lower rate of skipping was not a
result of an inherent difference in skipping probabilities in ASD, but is localised to this
BENCHMARK EFFECTS
39
region. Decreased skipping of the target word in the ASD group may reflect the adoption of a
more cautious reading strategy. It is possible that this is a result of the ASD readers being
sensitive to the manipulation of attachment and consequently the importance of the
disambiguating target noun. Therefore, decreased skipping may have been a strategy adopted
in order to avoid overlooking a potential misanalysis.
The second group difference was that ASD readers were found to have longer go past
times for the post-target region. Since this was present across both high and low attachment
conditions. If this was evidence of increased disruption for low attached sentences for readers
with ASD, we would have found an interaction whereby go past times were larger for ASD
readers for low attached sentences only. However, this is not the case and thus it is likely a
consequence of the increased re-reading that ASD participants engaged in for all sentences
(note that the post target region was often the penultimate word in the sentence).
Differences in the later stages of sentence processing were also detected. Replicating
what was found for Experiment 1, no group differences or interactions for first pass reading
were found. What this suggests is that the ASD readers did not differ from the TD group in
the speed with which they constructed an initial interpretation of the sentence. The ASD
group did differ however in the amount of second pass reading they engaged in, following
intact initial processing. Again this increased re-reading was found across both sentence types
and was unrelated to the sentence manipulation. As for Experiment 1, we speculate that this
may be a strategy adopted by the participants with ASD to ‘check’ their understanding of the
sentence, prior to (possibly) answering a comprehension question. Different reading
strategies have been previously identified in a TD population (Hyönä, Lorch & Kaakinen,
2002) and it is therefore possible that the increased re-reading and reduced skipping for the
target region may be part of a more general, cautious reading strategy, adopted by ASD
participants.
BENCHMARK EFFECTS
40
Finally, in relation to what would be expected in terms of typical cognitive processing
from general theories of reading, the overall findings from the two experiments for TD
readers replicate previous findings. In ASD readers however, this was not the case. Here, the
findings replicate those that would be predicted for normal cognitive processing in first pass
measures, but for second pass reading there was evidence for atypical cognitive processing
which is not in line with any general theory of reading.
Conclusion
We aimed to gain a more accurate understanding as to how lexical, syntactic and
semantic processing occurs in adults with ASD during natural reading, by measuring eye
movements as participants read single sentences. Overall, there were striking similarities in
first pass reading of sentences between TD and ASD readers. Experiment 1 indicated that
lexical processing is intact in ASD, with both groups showing a comparable frequency effect.
Experiment 2 indicated that there is no difference in syntactic preferences between the groups
of readers, and in their ability to detect and recover from an initial syntactic misanalysis
through the use of world knowledge. This was demonstrated by the immediacy with which an
initial syntactic misanalysis was detected and the magnitude of disruption to reading being
comparable between TD and ASD readers. More critically, what these findings imply, is that
an impairment in lexical identification, syntactic parsing or the use of to world knowledge
during the processing of single sentences is unlikely to contribute to the offline performance
differences for reading comprehension and inferencing tasks that have been reported in the
literature. The only group differences that were present in the current study appear to be
independent of the linguistic manipulations in our experimental sentences and, instead, reflect
a more cautious reading strategy that the ASD readers show a preference to adopt.
BENCHMARK EFFECTS
41
References
American Psychiatric Association. (2013). Diagnostic and statistical manual of mental
disorders (5th ed.). Arlington, VA: American Psychiatric Publishing.
Asberg, J., Kopp, S., Berg-Kelly, K., & Gillberg, C. (2010). Reading comprehension, word
decoding and spelling in girls with autism spectrum disorders (ASD) or attentiondeficit/hyperactivity disorder (AD/HD): performance and predictors. International
Journal of Language & Communication Disorders, 45(1), 61-71. doi:
10.3109/13682820902745438.
Au-Yeung, S. K., Kaakinen, J. K., Liversedge, S. P. & Benson, V. (2015). Processing of
Written Irony in Autism Spectrum Disorder: An Eye-Movement Study. Autism
Research. doi: 10.1002/aur.1490
Baayen, R. H., Davidson, D. G., & Bates, D. M. (2008). Mixed-effects modeling with crossed
random effects for subjects and items. Journal of Memory and Language, 59, 390412.
Baron-Cohen, S., Wheelwright, S., Skinner, R., Martin, J., & Clubley, E. (2001). The
Autism-Spectrum Quotient (AQ): Evidence from Asperger syndrome/highfunctioning autism, males and females, scientists and mathematicians. Journal of
Autism and Developmental Disorders, 31(1), 5-17. doi: 10.1023/a:1005653411471
Barr, D. J., Levy, R., Scheepers, C., & Tily, H. J. (2013). Random effects structure for
confirmatory hypothesis testing: Keep it maximal. Journal of Memory and Language,
68(3), 255-278. doi:10.1016/j.jml.2012.11.001
Bates D, Maechler M, Bolker B and Walker S (2014). lme4: Linear mixed-effects models
using Eigen and S4. R package version 1.1-7, http://CRAN.Rproject.org/package=lme4.
BENCHMARK EFFECTS
42
Brock, J. & Caruana, N. (2014). Reading for sound and reading for meaning in autism: Frith
and Snowling (1983) revisited. In Arciuli, J. & Brock, J (Eds.), Communication in
Autism (125-145). Amsterdam: John Benjamins Publishing Co.
Brock, J., & Bzishvili, S. (2013). Deconstructing Frith and Snowling's homograph-reading
task: Implications for autism spectrum disorders. The Quarterly Journal of
Experimental Psychology, 66(9), 1764-1773. doi:10.1080/17470218.2013.766221
Brock, J., Norbury, C., Einav, S., & Nation, K. (2008). Do individuals with autism process
words in context? Evidence from language-mediated eye-movements. Cognition,
108(3), 896-904. doi: 10.1016/j.cognition.2008.06.007.
Brysbaert, M., & New, B. (2009). Moving beyond Kucera and Francis: A critical evaluation
of current word frequency norms and the introduction of a new and improved word
frequency measure for American English. Behavior Research Measures, 41 (4), 977990. doi: 10.3758/BRM.41.4.977
Caruana, N., & Brock, B. (2014). No association between autisitic traits and contextual
influences on eye-movements during reading. PeerJ, 2:e446.
Coltheart, M., Rastle, K., Perry, C., Langdon, R., & Ziegler, J. (2001). DRC: a dual route
cascaded model of visual word recognition and reading aloud. Psychological Review,
108(1), 204.
Ehrlich, S. F., & Rayner, K. (1981). Contextual effects on word perception and eye
movements during reading. Journal of Verbal Learning and Verbal Behavior, 20(6),
641-655.
Eigsti, I. M., de Marchena, A. B., Schuh, J. M., & Kelley, E. (2011). Language acquisition in
autism spectrum disorders: A developmental review. Research in Autism Spectrum
Disorders, 5(2), 681-691. doi:10.1016/j.rasd.2010.09.001
BENCHMARK EFFECTS
43
Ferreira, F., Bailey, K. G., & Ferraro, V. (2002). Good-enough representations in language
comprehension. Current directions in psychological science, 11(1), 11-15.
Filik, R., & Moxey, L. M. (2010). The on-line processing of written irony. Cognition, 116(3),
421-436. doi:10.1016/j.cognition.2010.06.005
Frazier, L., & Clifton, C. (1996). Construal. MIT Press.
Frazier, L., & Rayner, K. (1982). Making and correcting errors during sentence
comprehension: Eye movements in the analysis of structurally ambiguous sentences.
Cognitive Psychology, 14(2), 178-210. doi: 10.1016/0010-0285(82)90008-1.
Frith, U. (1989). Autism: Explaining the Enigma. Oxford, UK: Basil Blackwell.
Frith, U., & Happé, F. (1994). Autism - Beyond the theory of mind. Cognition, 50(1-3), 115132.
Frith, U., & Snowling, M. (1983). Reading for meaning and reading for sound in autistic and
dyslexic children. British Journal of Developmental Psychology, 1, 329-342.
Graesser, A. C., Singer, M., & Trabasso, T. (1994). Constructing inferences during narrative
text comprehension. Psychological Review, 101(3), 371-395.
Grainger, J., & Jacobs, A. M. (1996). Orthographic processing in visual word recognition: a
multiple read-out model. Psychological review, 103(3), 518.
Hahn, N., Snedeker, J., & Rabagliati, H. (2015). Rapid linguistic ambiguity resolution in
young children with autism spectrum disorder: Eye tracking evidence for the limits of
weak central coherence. Autism Research. doi: 10.1002/aur.1487
Hala, S., Pexman, P. M., & Glenwright, M. (2007). Priming the meaning of homographs in
typically developing children and children with autism. Journal of Autism and
Developmental disorders, 37(2), 329-340. doi: 10.1007/s10803-006-0162-6
BENCHMARK EFFECTS
44
Happé, F. G. E. (1997). Central coherence and theory of mind in autism: Reading
homographs in context. British Journal of Developmental Psychology, 15, 1-12.
doi: 10.1111/j.2044-835X.1997.tb00721.x
Happé, F., & Frith, U. (2006). The weak coherence account: detail-focused cognitive style in
autism spectrum disorders. Journal of Autism and Developmental Disorders, 36(1), 525. doi: 10.1007/s10803-005-0039-0.
Henderson, L. M., Clarke, P. J., & Snowling, M. J. (2011). Accessing and selecting word
meaning in autism spectrum disorder. Journal of Child Psychology and Psychiatry,
52(9), 964-973. doi: 10.1111/j.1469-7610.2011.02393.x
Huemer, S. V., & Mann, V. (2010). A comprehensive profile of decoding and comprehension
in autism spectrum disorders. Journal of Autism and Developmental Disorders, 40(4),
485-493. Doi: 10.1007/s10803-009-0892-3.
Hyönä, J., Lorch Jr, R. F., & Kaakinen, J. K. (2002). Individual differences in reading to
summarize expository text: Evidence from eye fixation patterns. Journal of
Educational Psychology, 94(1), 44-55. DOI: 10.1037//0022-0663.94.1.44
Inhoff, A. W., & Rayner, K. (1986). Parafoveal word processing during eye fixations in
reading: Effects of word frequency. Perception & Psychophysics, 40(6), 431-439.
Jeffreys, H. (1961). Theory of probability (3rd ed.). New York: Oxford University Press.
Jolliffe, T., & Baron-Cohen, S. (1999). A test of central coherence theory: linguistic
processing in high-functioning adults with autism or Asperger syndrome: is local
coherence impaired? Cognition, 71(2), 149-185.
Jolliffe, T., & Baron-Cohen, S. (2000). Linguistic processing in high-functioning adults with
autism or Asperger's syndrome. Is global coherence impaired? Psychological
Medicine, 30(5), 1169-1187.
BENCHMARK EFFECTS
45
Jones, C. R. G., Happé, F., Golden, H., Marsden, A. J. S., Tregay, J., Simonoff, E., . . .
Charman, T. (2009). Reading and Arithmetic in Adolescents With Autism Spectrum
Disorders: Peaks and Dips in Attainment. Neuropsychology, 23(6), 718-728. doi:
10.1037/a0016360.
Joseph, H. S. S. L., & Liversedge, S. P. (2013). Children’s and adults’ on-line processing of
syntactically ambiguous sentences during reading. PloS one, 8(1), e54141.
Kamio, Y., & Toichi, M. (2000). Dual access to semantics in autism: Is pictorial access
superior to verbal access?. Journal of Child Psychology and Psychiatry, 41(07), 859867. doi:http://dx.doi.org/
Kamio, Y., Robins, D., Kelley, E., Swainson, B., & Fein, D. (2007). Atypical
lexical/semantic processing in high-functioning autism spectrum disorders without
early language delay. Journal of Autism and Developmental Disorders, 37(6), 11161122. doi: 10.1007/s10803-006-0254-3. doi:10.1016/j.clinph.2013.06.021
Kass, R. E., & Raftery, A. E. (1995). Bayes factors. Journal of the American Statistical
Association, 90(430), 773-795.
Koolen, S., Vissers, C. T. W., Egger, J. I. M., & Verhoeven, L. (2014). Monitoring in
language perception in high-functioning adults with autism spectrum disorder:
Evidence from event-related potentials. Clinical Neurophysiology, 125(1), 108-123.
Koolen, S., Vissers, C. T. W., Hendriks, A. W., Egger, J. I., & Verhoeven, L. (2012). The
interplay between attentional strategies and language processing in high-functioning
adults with autism spectrum disorder. Journal of Autism and Developmental
Disorders, 42(5), 805-814. doi:10.1007/s10803-011-1310-1
Lopez, B., & Leekam, S. R. (2003). Do children with autism fail to process information in
context? Journal of Child Psychology and Psychiatry and Allied Disciplines, 44(2),
285-300. doi:10.1111/1469-7610.00121
BENCHMARK EFFECTS
46
Lord C., Rutter M., DiLavore P. C., Risi S., Gotham K., Bishop S. (2012). Autism diagnostic
observation schedule, second edition. Torrance, CA: Western Psychological Services.
Lucas, R., & Norbury, C. F. (2014). Levels of text compehension in children with autism
spectrum disorders (ASD): The influence of language phenotype. Journal of Auitsm
and Developmental Disorders, 44, 2756-2768. doi:10.1007/s10803-014-2133-7
MacDonald, M.C., Pearlmutter, N.J., & Seidenberg, M.S. (1994). The lexical nature of
syntactic ambiguity resolution. Psychological Review, 101, 676-703.
Mayes, S. D., & Calhoun, S. L. (2006). Frequency of reading, math, and writing disabilities
in children with clinical disorders. Learning and Individual Differences, 16(2), 145157. doi: 10.1016/j.lindif.2005.07.004
McKoon, G., & Ratcliff, R. (1992). Inference during reading. Psychological Review, 99(3),
440-466.
Minshew, N. J., & Goldstein, G. (1998). Autism as a disorder of complex information
processing. Mental Retardation and Developmental Disabilities Research Reviews,
4(2), 129-136.
Minshew, N. J., Goldstein, G., & Siegel, D. J. (1995). Speech and language in highfunctioning autistic individuals. Neuropsychology, 9(2), 255-261.
Minshew, N. J., Goldstein, G., & Siegel, D. J. (1997). Neuropsychologic functioning in
autism: profile of a complex information processing disorder. Journal of the
International Neuropsychological Society : JINS, 3(4), 303-316.
Minshew, N. J., Williams, D. L., & McFadden, K. (2008). Information Processing, Neural
Connectivity, and Neuronal Organization. In A. W. Zimmerman (Ed.), Autism:
Current Theories and Evidence (pp. 381-405).
BENCHMARK EFFECTS
47
Nation, K., Clarke, P., Wright, B., & Williams, C. (2006). Patterns of reading ability in
children with autism spectrum disorder. Journal of Autism and Developmental
Disorders, 36(7), 911-919. doi:10.1007/s10803-006-0130-1
Norbury, C. F. (2005). Barking up the wrong tree? Lexical ambiguity resolution in children
with language impairments and autistic spectrum disorders. Journal of Experimental
Child Psychology, 90(2), 142-171. doi:10.1016/j.jecp.2004.11.003
R Core Team (2015). R: A language and environment for statistical computing. R Foundation
for Statistical Computing, Vienna, Austria. URL https://www.R-project.org/.
Rayner, K. (1998). Eye movements in reading and information processing: 20 years of
research. Psychological Bulletin, 124(3), 372. doi: org/10.1037/0033-2909.124.3.372.
Rayner, K. (2009). Eye movements and attention in reading, scene perception, and visual
search. Quarterly Journal of Experimental Psychology, 62(8), 1457-1506.
doi:10.1080/17470210902816461
Rayner, K., & Duffy, S. A. (1986). Lexical complexity and fixation rimes in reading – Effects
of word frequency, verb complexity, and lexical ambiguity. Memory & Cognition, 14
(3), 191-201. doi:10.3758/bf03197692
Rayner, K., Carlson, M., & Frazier, L. (1983). The interaction of syntax and semantics during
sentence processing: Eye movements in the analysis of semantically biased sentences.
Journal of Verbal Learning and Verbal Behavior, 22, 358-374.
Riches, N. G., Loucas, T., Baird, G., Charman, T., & Simonoff, E. (2015). Elephants in
Pyjamas: Testing the Weak Central Coherence Account of Autism Spectrum
Disorders Using a Syntactic Disambiguation Task. Journal of Autism and
Developmental Disorders, 1-9. doi:10. 1007/ s10803-015-2560-0
Rouder, J. N., Morey, R. D., Speckman, P. L., & Province, J. M. (2012). Default Bayes
factors for ANOVA designs. Journal of Mathematical Psychology, 56, 365-374.
BENCHMARK EFFECTS
48
doi:10.1016/j.jmp.2012.08.001
Rumelhart, D. E., & McClelland, J. L. (1982). An interactive activation model of context
effects in letter perception: II. The contextual enhancement effect and some tests and
extensions of the model. Psychological Review, 89(1), 60.
Saldaña, D., & Frith, U. (2007). Do readers with autism make briging inferences from world
knowledge? Journal of Experimental Child Psychology, 96, 310-319.
doi:10.1016/j.jecp.2006.11.002.
Saldaña, D., Carreiras, M., & Frith, U. (2009). Orthographic and Phonological Pathways in
Hyperlexic Readers With Autism Spectrum Disorders. Developmental
Neuropsychology, 34(3), 240-253. doi:10.1080/87565640902805701
Sansosti, F. J., Was, C., Rawson, K. A., & Remaklus, B. L. (2013). Eye movements during
processing of text requiring bridging inferences in adolescents with higher functioning
autism spectrum disorders: A preliminary investigation. Research in Autism Spectrum
Disorders, 7(12), 1535-1542. doi:10.1016/j.rasd.2013.09.001
Seidenberg, M. S., & McClelland, J. L. (1989). A distributed, developmental model of word
recognition and naming. Psychological Review, 96(4), 523.
Semel, E., Wiig, E. H., & Secord, W. A. (2003). Clinical evaluation of language
fundamentals, fourth edition (CELF-4). Toronto, Canada: The Psychological
Corporation/A Harcourt Assessment Company.
Snowling, M.J., Stothard, S.E., Clarke, P., Bowyer-Crane, C., Harrington, A., Truelove, E.,
Nation, K. & Hulme, C. (2010). York Assessment of Reading for Comprehension:
Passage Reading - Secondary Version. GL Assessment.
Stockbridge, M. D., Happé, F. G. E., & White, S. J. (2014). Impaired comprehension of
altering syntactic constructions in autism. Autism Research, 7, 314-321.
BENCHMARK EFFECTS
49
Taraban, R., & McClelland, J. L. (1988). Constituent attachment and thematic role
assignment in sentence processing: Influences of content based expectations. Journal
of Memory and Language, 27(6), 597-632. doi:10.1016/0749-596x(88)90011-3
Toichi, M., & Kamio, Y. (2001). Verbal association for simple common words in highfunctioning autism. Journal of Autism and Developmental Disorders, 31(5), 483-490.
doi: 10.1023/A:1012216925216
Van Gompel, R. P., Pickering, M. J., & Traxler, M. J. (2000). Unrestricted race: A new
model of syntactic ambiguity resolution. Reading as a perceptual process, 621-648.
Venables, W. N. & Ripley, B. D. (2002) Modern Applied Statistics with S. Fourth Edition.
Springer, New York. ISBN 0-387-95457-0
Wagenmakers, E. (2007). A practical solution to the pervasive problems of p values.
Psychonomic Bulletin & Review, 14(5), 779-804.
Williams, Diane, L., Goldstein, G., & Minshew, N. J. (2006). Neuropsychologic functioning
in children with autism: Further evidence for disordered complex informationprocessing. Child Neuropsychology, 12, 279-298. Doi: 10.1080/09297040600681190
Wahlberg, T., & Magliano, J. P. (2004). The ability of high function individuals with autism
to comprehend written discourse. Discourse Processes, 38(1), 119-144.
doi:10.1207/s15326950dp3801_5
Wechsler, D. (1999). Wechsler Abbreviated Scale of Intelligence. The Psychological
Corporation: Harcourt Brace & Company. New York, NY.
BENCHMARK EFFECTS
Table 1.
Model parameters and observed means for Experiment 1 global analysis.
Model
TD
b
SE
t
Sig
High
Low
Intercept
5.40 0.02 279.08 *
50
ASD
High
Low
-0.05 0.04 -1.24
MFD Group
217 (30)
220 (30)
229 (40)
233 (41)
Frequency
0.02 0.01 2.54
*
Group X Frequency
< 0.01 0.02 -0.29
Intercept
2.40 0.05 44.83 *
Group
-0.24 0.10 -2.45
*
MFC
10 (4)
11 (4)
13 (6)
14 (6)
Frequency
0.02 0.02 1.33
Group X Frequency
0.03 0.03 0.86
Intercept
7.81 0.06 126.81 *
Group
-0.28 0.12 -2.41
*
SRT
2243 (905) 2359 (971) 3174 (1864) 3223 (1723)
Frequency
0.03 0.02 1.81
Group X Frequency
0.03 0.03 0.93
Intercept
6.37 0.11 57.46 *
Group
-0.47 0.22 -2.11
*
SSRT
774 (727) 679 (624) 1483 (1615) 1555 (1574)
Frequency
0.01 0.07 0.09
Group X Frequency
-0.26 0.13 -1.91
Nb. MFD = mean fixation duration, MFC = mean fixation count, SRT = sentence reading time, SSRT = second
pass sentence reading time.
BENCHMARK EFFECTS
51
Table 2.
Model parameters for Experiment 1 region analyses.
Start
Target
b
SE
t
Sig
b
SE
Intercept
-2.57 0.23
Group
0.67 0.41
SKIP† Frequency
-0.59 0.32
Group X
0.23 0.48
Frequency
Intercept
5.44 0.03
Group
-0.03 0.05
FFD Frequency
0.08 0.02
Group X
0.04 0.04
Frequency
Intercept
5.48 0.03
Group
-0.05 0.05
SFD Frequency
0.11 0.03
Group X
0.01 0.05
Frequency
Intercept
6.86 0.08 82.42 *
5.62 0.04
Group
-0.11 0.11 -1.03
-0.04 0.07
GD
Frequency
0.00 0.02 -0.12
0.17 0.03
Group X
0.03 0.04 0.75
0.03 0.05
Frequency
Intercept
5.86 0.05
Group
-0.18 0.09
TT
Frequency
0.23 0.05
Group X
-0.09 0.08
Frequency
Intercept
6.00 0.09 63.49 *
5.55 0.06
Group
-0.38 0.18 -2.10
*
-0.16 0.12
SPT Frequency
0.02 0.07 0.23
0.15 0.08
Group X
-0.26 0.14 -1.91
0.04 0.15
Frequency
Intercept
0.38 0.28 1.38
-1.52 0.12
Group
-0.85 0.53 -1.62
-0.22 0.21
RI†
Frequency
-0.21 0.20 0.28
0.39 0.21
Group X
0.23 0.29 0.80
0.39
Frequency
13.00
Intercept
-1.79 0.23
Group
-0.30 0.33
RO† Frequency
-0.11 0.18
Group X
-0.42 0.35
Frequency
†
Values in the t column for this variable correspond to z values.
End
t
Sig
-11.33 *
1.63
-1.83
b
SE
t
Sig
0.48
203.20
-0.56
4.02
*
*
1.03
180.78
-0.86
4.45
*
*
0.26
151.52
-0.64
6.03
*
*
0.57
119.92
-2.02
4.86
6.20 0.08 75.67
-0.07 0.11 -0.60
0.03 0.04 0.71
0.10 0.07
*
1.35
*
*
*
-1.08
94.29
-1.36
1.85
*
0.26
-12.63
-1.03
1.88
6.07 0.09 67.75
-0.50 0.16 -3.20
-0.13 0.09 -1.57
*
*
-0.09 0.16 -0.56
*
-0.33
-7.67
-0.92
-0.59
-1.22
*
0.58 0.27 2.15
-0.88 0.53 -1.67
-0.02 0.17 -0.15
0.55 0.34
*
1.65
Nb. SKIP = skipping, FFD = first fixation duration, SFD = single fixation duration, GD = gaze duration, TT
= total time, SPT = second pass time, RI = regressions in, RO = first pass regressions out.
BENCHMARK EFFECTS
52
Table 3.
Observed means (standard deviations) for Experiment 1 region analyses.
SKIP
FFD
SFD
GD
TT
SPT
RI
RO
High
1025 (525)
504 (417) .49 (.50)
TD
Low
1028 (497)
481 (427) .47 (.50)
Start
High
1172 (613)
869 (837) .63 (.48)
ASD
Low
1188 (707)
960 (874) .59 (.49)
High .14 (.35) 225 (67) 230 (75) 268 (129) 334 (184) 264 (180) .16 (.37) .19 (.40)
TD
Low .12 (.32) 253 (81) 260 (83) 325 (147) 400 (212) 321 (256) .20 (.40) .15 (.36)
Target
High .09 (.29) 239 (84) 245 (88) 281 (114) 385 (231) 386 (469) .18 (.38) .21 (.40)
ASD
Low .07 (.26) 257 (94) 274 (97) 335 (154) 522 (338) 422 (353) .24 (.43) .23 (.42)
High
569 (372)
492 (342)
.49 (.50)
TD
Low
611 (393)
441 (444)
.55 (.50)
End
High
679 (485)
948 (815)
.65 (.48)
ASD
Low
648 (440)
845 (657)
.63 (.48)
Nb. SKIP = skipping, FFD = first fixation duration, SFD = single fixation duration, GD = gaze duration, TT
= total time, SPT = second pass time, RI = regressions in, RO = first pass regressions out.
BENCHMARK EFFECTS
53
Table 4.
Model parameters and observed means for Experiment 2 global analysis.
Model
TD
ASD
b
SE
t
Sig
High
Low
High
Low
Intercept
5.39 0.02 274.52 *
Group
-0.03 0.04 -0.80
< 0.01 0.01 0.31
MFD Frequency
218 (31)
219 (30)
226 (36)
226 (38)
Group X
0.01 0.01 0.48
Frequency
Intercept
2.67 0.06 47.69 *
Group
-0.33 0.11 -2.98
*
0.09 0.02 5.33
*
MFC Frequency
12 (4)
14 (5)
18 (9)
20 (11)
Group X
-0.01 0.04 -0.37
Frequency
Intercept
8.07 0.07 123.41 *
Group
-0.36 0.13 -2.78
*
0.09 0.02 4.39
*
SRT Frequency
2739 (1069) 3015 (1277) 4171 (2486) 4687 (2898)
Group X
-0.01 0.04 -0.29
Frequency
Intercept
6.61 0.13 50.85 *
Group
-0.66 0.26 -2.56
*
0.27 0.06 4.73
*
SSRT Frequency
774 (715) 1064 (959) 2064 (2330) 2571 (2679)
Group X
0.08 0.11 0.68
Frequency
Nb. MFD = mean fixation duration, MFC = mean fixation count, SRT = sentence reading time, SSRT =
second pass sentence reading time.
BENCHMARK EFFECTS
Table 5.
Model parameters for Experiment 2 start, verb, noun and preposition region analyses.
Start
Verb
b
SE
t
Sig
b
SE
t
Sig
b
Intercept
*
*
5.72
0.05
107.85
5.53
0.04 125.71
6.08
Group
-0.15
0.09
-1.63
-0.08
0.08
-1.02
-0.13
GD Frequency
0.03
0.02
1.33
< 0.01 0.02
0.24
-0.03
Group X
0.03
0.04
0.68
< 0.01 0.04
0.12
0.05
Frequency
Intercept
*
*
5.61
0.07
78.10
5.72
0.06
92.71
5.92
Group
*
-0.33
0.14
-2.47
-0.23 -0.12
-1.97
-0.41
SPT Frequency
-0.06
0.07
-0.89
0.06
0.06
1.08
0.18
Group X
-0.07
0.13
-0.55
0.08
0.11
0.72
0.07
Frequency
Intercept
*
*
-1.40
0.27
-5.23
-0.81
0.26
-.3.12
-0.94
Group
-0.86
0.50
-1.71
-0.63
0.51
-1.22
-0.77
RI†
Frequency
0.17
0.15
1.13
0.09
0.14
0.65
0.34
Group X
-0.47
0.26
-1.78
0.32
0.27
1.21
0.26
Frequency
Intercept
*
-2.56
0.20 -12.57
-2.07
Group
0.27
0.39
0.68
0.02
RO† Frequency
-0.03
0.18
-0.17
0.07
Group X
-0.44
0.37
-1.19
0.16
Frequency
†
Note that the values in the t column for this variable correspond to z values.
Nb. GD = gaze duration, SPT = second pass time, RI = regressions in, RO = first pass regressions out.
54
Noun
SE
t
0.06 103.02
0.10
-1.38
0.03
-0.83
0.06
0.82
0.09
0.16
0.07
68.83
-2.57
2.40
0.16
0.44
0.24
0.47
0.16
-3.90
-1.64
2.08
0.32
0.82
0.18
0.34
0.18
-11.26
0.07
0.39
0.30
0.53
Sig
*
*
*
*
*
*
*
b
5.31
-0.03
0.03
Preposition
SE
t
0.04 146.75
0.07
-0.48
0.03
0.95
< 0.01
0.05
0.19
5.60
-0.16
0.14
0.07
0.15
0.10
75.76
-1.03
1.39
0.08
0.21
0.38
-0.69
-0.31
0.55
0.21
0.40
0.16
-3.27
-0.78
3.47
0.07
0.31
0.22
-2.45
-0.21
0.03
0.21
0.35
0.32
-11.58
-0.61
0.10
-0.23
0.51
-0.45
Sig
*
*
*
*
*
BENCHMARK EFFECTS
55
Table 6.
Model parameters for Experiment 2 pre-target, target, post target and end regions.
Pre-target
b
SE
t
Sig
-3.25
0.29
-10.96
*
Intercept
Group
0.33
0.46
0.72
†
SKIP
0.58
0.35
1.65
Frequency
Group X Frequency
-0.02
0.52
-0.05
5.29
0.19
273.44
*
Intercept
Group
0.06
0.04
1.50
FFD
-0.02
0.01
-1.25
Frequency
0.01
0.03
0.43
Group X Frequency
SFD
GD
GP
TT
SPT
RI†
†
RO
Intercept
Group
Frequency
Group X Frequency
Intercept
Group
Frequency
Group X Frequency
Intercept
Group
Frequency
Group X Frequency
Intercept
Group
Frequency
Group X Frequency
Intercept
Group
Frequency
Group X Frequency
Intercept
Group
Frequency
Group X Frequency
Intercept
Group
Frequency
Group X Frequency
b
-2.27
0.82
-0.25
0.13
5.42
0.02
0.02
Target
SE
t
0.25
-9.21
0.39
2.13
0.19
-1.28
0.33
0.38
0.03
215.70
0.05
0.42
0.02
1.24
-0.01
0.03
-0.40
5.46
0.02
0.06
0.02
0.03
0.06
0.02
0.04
186.04
0.29
2.49
0.41
*
5.58
-0.01
0.09
< 0.01
0.03
0.06
0.03
0.04
172.10
-0.17
3.00
0.02
*
5.75
-0.07
0.10
0.01
0.04
0.08
0.03
0.05
130.58
-0.79
3.07
0.21
*
5.33
0.02
-0.01
0.02
0.02
0.04
0.02
0.03
233.44
0.45
-0.73
0.53
*
5.53
-0.09
0.01
-0.03
0.03
0.06
0.02
0.05
166.66
-1.52
0.60
-0.70
*
5.68
-0.18
-0.02
-0.01
0.05
0.10
0.03
0.04
108.09
-1.79
-0.82
-0.19
*
5.96
-0.33
0.11
-0.01
0.07
0.13
0.04
0.07
90.57
-2.62
2.52
-0.10
*
*
*
5.82
-0.18
0.18
-0.04
0.05
0.10
0.04
0.05
107.04
-1.73
4.41
-0.85
5.75
-0.35
0.15
0.14
0.07
-0.49
0.43
0.11
-2.30
-0.61
-0.15
0.08
0.06
0.13
0.07
0.13
0.18
0.35
0.12
0.23
0.22
0.41
0.19
0.32
90.30
-2.76
2.32
1.15
-4.07
-1.39
3.57
0.46
-10.30
-1.46
-0.79
0.26
*
*
*
5.62
-0.33
0.13
0.11
-1.40
-0.12
0.31
-0.08
-1.72
-0.15
0.12
-0.10
0.06
0.11
0.06
0.13
0.17
0.28
0.14
0.27
0.16
0.32
0.17
0.30
94.93
-2.95
1.95
0.84
-8.90
-0.42
2.15
-0.28
-10.57
-0.46
0.69
-0.33
*
*
*
Sig
*
*
*
*
*
*
*
*
*
*
*
*
b
-1.67
0.09
-0.29
0.21
5.40
-0.02
0.02
Post Target
SE
t
0.36
-4.64
0.41
0.23
0.20
-1.44
0.27
0.78
0.03
200.86
0.05
-0.29
0.02
0.94
End
-0.03
0.04
-0.79
5.44
-0.04
< 0.01
-0.03
0.03
0.06
0.03
0.06
176.97
-0.66
0.03
-0.53
*
5.62
-0.10
0.01
0.05
0.05
0.08
0.03
0.06
124.41
-1.18
0.42
0.77
*
60.40
-0.29
0.15
0.05
0.07
0.13
0.05
0.10
85.15
-2.13
2.86
0.50
*
*
*
5.90
-0.30
0.08
0.03
0.06
0.12
0.03
0.07
94.07
-2.63
2.32
0.37
*
*
*
5.69
-0.25
0.06
-0.11
-1.19
-0.43
0.04
0.28
-0.95
-0.02
0.40
-0.03
0.07
0.14
0.08
0.16
0.17
0.25
0.14
0.26
0.18
0.28
0.13
0.25
83.54
-1.83
0.74
-0.71
-7.16
-1.72
0.32
1.07
-5.34
-0.06
3.07
-0.13
*
Sig
*
b
SE
t
Sig
5.62
-0.01
-0.03
-0.06
0.05
0.08
0.03
0.06
110.62
-0.13
-0.95
-1.07
*
5.86
-0.55
0.08
0.04
0.07
0.14
0.09
0.17
83.08
-3.90
0.94
0.24
*
*
0.81
-0.77
0.23
0.21
0.25
0.49
0.16
0.30
3.22
-1.56
1.43
0.68
*
*
*
*
*
† Note that the values in the t column for this variable correspond to z values.
Nb. SKIP = skipping, FFD = first fixation duration, SFD = single fixation duration, GD = gaze duration, TT = total time, SPT = second pass time, RI = regressions in, RO = first pass regressions out.
BENCHMARK EFFECTS
56
Table 7.
Observed means (standard deviations) for Experiment 2 region analyses.
SKIP
FFD
SFD
GD
TD
Start
ASD
TD
Verb
ASD
TD
Noun
ASD
Prep
TD
ASD
TD
Pre
target
ASD
TD
Target
ASD
TD
Post
Target
ASD
GP
TT
SPT
RI
RO
High
Low
High
Low
304 (137)
325 (196)
357 (167)
364 (176)
294 (178)
266 (159)
490 (403)
542 (623)
.21 (.41)
.20 (.40)
.31 (.47)
.38 (.49)
High
Low
High
Low
268 (122)
272 (129)
297 (126)
301 (139)
321 (204)
368 (218)
466 (333)
527 (513)
.26 (.44)
.32 (.47)
.42 (.49)
.41 (.50)
.11 (.32)
.10 (.30)
.07 (.26)
.09 (.29)
High
Low
High
Low
469 (261)
472 (259)
564 (302)
553 (332)
405 (315)
495 (389)
773 (714)
864 (751)
.23 (.42)
.31 (.46)
.37 (.48)
.41 (.49)
.14 (.34)
.15 (.36)
.15 (.36)
.14 (.35)
High
Low
High
Low
223 (105)
229 (91)
236 (106)
241 (118)
259 (144)
328 (202)
364 (253)
447 (327)
.29 (.45)
.40 (.49)
.09 (.28)
.09 (.28)
.38 (.49)
.47 (.50)
.09 (.29)
.11 (.32)
High
Low
High
Low
.07 (.26)
.09 (.28)
.07 (.26)
.08 (.27)
212 (50)
209 (48)
203 (57)
198 (52)
216 (51)
214 (48)
212 (63)
210 (73)
261 (115)
262 (117)
288 (138)
299 (144)
307 (164)
299 (167)
381 (269)
373 (234)
356 (197)
408 (236)
564 (485)
616 (447)
291 (179)
373 (233)
546 (439)
582 (518)
.27 (.44)
.36 (.48)
.37 (.48)
.44 (.50)
.11 (.32)
.10 (.30)
.20 (.40)
.17 (.37)
High
Low
High
Low
.22 (.41)
.19 (.39)
.13 (.34)
.10 (.30)
239 (84)
243 (82)
234 (82)
243 (90)
242 (88)
259 (98)
242 (87)
254 (93)
276 (116)
300 (119)
278 (112)
305 (127)
328 (178)
364 (198)
375 (390)
407 (306)
322 (160)
386 (203)
401 (277)
503 (355)
253 (144)
315 (193)
445 (389)
474 (383)
.22 (.41)
.26 (.44)
.20 (.40)
.25 (.43)
.16 (.37)
.18 (.38)
.18 (.38)
.21 (.41)
High
Low
High
Low
.27 (.44)
.25 (.43)
.25 (.43)
.21 (.41)
236 (97)
237 (95)
236 (90)
248 (107)
244 (105)
244 (105)
250 (104)
253 (107)
298 (154)
312 (175)
339 (180)
340 (201)
421 (356)
541 (472)
684 (816)
890 (1256)
361 (210)
391 (221)
533 (381)
570 (415)
311 (235)
321 (281)
482 (378)
565 (522)
.22 (.42)
.25 (.43)
.32 (.47)
.31 (.46)
.27 (.45)
.35 (.48)
.28 (.45)
.35 (.48)
High
349 (205)
345 (314) 674 (505)
.54 (.50)
Low
324 (199)
338 (204) 644 (452)
.60 (.49)
End
High
342 (227)
.67 (.47)
ASD
Low
349 (236)
.72 (.45)
Nb. Prep = preposition, SKIP = skipping, FFD = first fixation duration, SFD = single fixation duration, GD = gaze duration, TT = total time, SPT
= second pass time, RI = regressions in, RO = first pass regressions out.
TD
BENCHMARK EFFECTS
57
Acknowledgements
This research was funded by the Mayflower scholarship, provided by the Faculty of
Health, Life and Mathematical Sciences at the University of Southampton and was completed
in part for the first author’s doctoral thesis. We would like to thank all of the postgraduates
from the Centre of Vision and Cognition who assisted with data collection and all of the
volunteers who participated.